Transcription
An encrypted message or a cipher presents us with a strange paradox: a deeply unsettling repulsion and yet, at the same time, an eroticism of representation. Its message unflinchingly stares out to us, and yet, as we stare back into it, otherwise familiar letters and numbers refuse to yield their underlying meaning. We're invited to peer into the message, and yet the cipher refuses to yield. That it holds the secret is all the more enticing. Indeed, the only thing worse than a cipher is a solution to that cipher. The mystery evaporates, and we're left a bit bereft, having traded the allure of the esoteric for the plainness of everyday meaning.
Of course, adding centuries of intrigue and the trappings of the occult only deepened that allure, and further surrounding that cipher with equally alien vistas of unknown flora, astrological wheels, seeming glimpses of the cosmos itself, scores of naked nymphs and interlocking baths, and vast lists of apparent pharmacological lore only solidifies the seductive power of that mystery. Every single page seems bursting with meaning, and yet, as we pour over page after page, the mystery only deepens, and we find ourselves fully in the grips of a secret that seems so close and yet remains so far away.
The apparent medieval cipher of the Voynich Manuscript, filled with hundreds of pages of such mysteries, is the paragon of just such a seductive enigma. Though in the past two decades, truly contemporary cryptology, statistics, and information theory, along with a host of other material sciences, have been brought to bear in the desperate hopes of a solution for just that text. But are we any closer to unlocking the mystery of the Voynich Manuscript?
In this episode, I want to explore what is increasingly becoming a viable position on that question: that the Voynich Manuscript itself does hide a secret. But that secret isn't a meaning hidden behind elaborate encryption. Rather, the secret is that the founding manuscript itself may well be a medieval hoax. Using robust computational analysis, statistics, and information theory, scientists and scholars may be inching their way to unlocking just how an apparent medieval cipher is actually generated using a quasi-algorithmic process that, while incredibly simple, produces effects that have baffled cryptologists for centuries.
If you're interested in magic, hermetic philosophy, alchemy, Kabbalah, or the history of the occult, make sure to subscribe here to Esoterica and check out my other topics on contents and esotericism, including curated playlists. Also, if you want to support my work of providing accessible, scholarly, and free content on topics and esotericism here on YouTube for free, I'd hope you consider taking a look at my Patreon. Perhaps consider a one-time donation to the channel, or you can even use the nifty Super Thanks option just below the video if you want to do that. But now, let's turn to cutting-edge research on the Voynich Manuscript and how its meaning has proved elusive precisely because only the appearance of meaning was perhaps ever there to begin with. That it may well be an elaborate medieval hoax. I'm Dr. Justin Sledge, and welcome to Esoterica, where we explore the arcane in history, philosophy, and religion.
[Music]
[Music]
Typically, people interested in topics and esotericism find that over time, those interests tend to broaden. From an interest in astrology may come an interest in medieval medicine. From alchemy to symbolism more generally. From magic in the contemporary period to the theory of its mechanics and Renaissance philosophy. As you probably can imagine, many of those roads will eventually lead to an interest in hidden meanings more generally, and the Voynich Manuscript stands at the crossroads of most of those interests.
Of course, I've done episodes on the origins of magical symbols in Greco-Roman Egypt, the quasi-constructed language and alphabet of Hildegard of Bingen's Lingua Ignota. There are whole playlists that I've produced on alchemy and its symbolism, the algorithmically produced book of Soy Had by Dr. John Dee, and of course, a whole series on the so-called Enochian alphabet and language revealed to Dee and Kelley. My interest, and likely your interest, in the Voynich Manuscript should come as no surprise to anyone. Even as a teenager, I volunteered on the European Voynich Manuscript transcription team. I carried my black and white photostat copy around in my car so I could work on it if I ever got spare time. I even got that text from Yale for my like 17th birthday, before any facsimiles were being published or online images were available at all.
As a dedicated amateur, I banged my head against the mystery for some years, making no more headway than anyone else. I'm sure less headway than certain experts. And as the new millennium rolled in, the major transcriptions into machine-readable code were finished. And as far as I'm concerned, the matter would be worked out in a few years under the brute force of modern computational analysis and cryptography. There was just no way any of us were going to compete by hand. But in the two decades that have followed, the Voynich is still no closer to decryption. In fact, the informational mystery has only gotten deeper.
But before proceeding, I'll just say that this episode is not meant to be a primer or an introductory episode on the Voynich Manuscript. I'm going to assume some basic familiarity with the text, its history, and previous attempts at decipherment. But I've included a link in the card above for a great primer by the Histocrat if you really want to review that first to get into the world of Voynicholi. But briefly, the Voynich Manuscript, or more properly Yale Beinecke Manuscript 408, is an apparent medieval cipher manuscript of about 240 pages. It is heavily illustrated and is usually broken down into an herbal section, astronomical and biological sections – that's the section with those strange naked nymphs swimming around in pipes that are kind of vaguely gynecological – there's a cosmological section, a pharmaceutical section, and just a rough recipe section.
The herbal section spans more than half of the totality of the manuscript at 112 folios, though the other sections are often the most densely packed with Voynich text. Voynichese, as we call it, it contains over 170,000 characters of an alphabet that's roughly 20 to 25 characters in totality, gathered into apparently 28,000 fluently written words. Words. There's no punctuation, and stranger still, there's virtually no internal corrections to the text, about 12, maybe total 12. The codex is incomplete as we have it. There are some sections that were removed prior to the text being acquired by the book dealer Wilfred Voynich in 1912. Since 1969, it's been housed at the Yale University Beinecke Rare Book and Manuscript Library, which is an amazing place.
Carbon dating of the vellum places its production between 1404 and 1438 with about 95% accuracy on that reading. The script and illustrations are consistent with that dating. It first enters into history sometime in the late 16th, early 17th century in the court of the mercurial Holy Roman Emperor Rudolph II, because of course it does. From there, it would make its way to Rome in the hopes of a decipherment by the famed scholar Athanasius Kircher. Following his death, it was held by the Jesuits in the Collegio Romano until it was purchased by Voynich. Despite a century of modern attempts, the text remains undeciphered, hence it being a topic on Esoterica. If it were deciphered, it would be a topic on Exoterica.
At this point, the general cryptological consensus is that the Voynich is either an unknown natural language, a known natural language has been encrypted, a constructed language that's either encrypted or not, and a possible non-language meant to look like a language. It's very unlikely it's a known natural language, or even a known natural language in some form of medieval cryptography or medieval encryption. Frankly, computational and cryptological analysis would have very likely revealed that by now, and such analysis contra-indicates that possibility, especially from a statistical point of view.
Now, it could be an exotic natural language, that is to say, a non-native language of the region in which the manuscript was produced, written in a kind of custom-made script for that language at that time. The problem here is that computational analysis of the manuscript reveals properties that no known natural language of the 57, 58, 59, 100 or so natural known languages in the world seem to have. It has weird properties unknown in any natural language. For instance, Voynichese, as it's called, does obey Zipf's laws, revealing that it was very likely produced by a method that was not purely stochastic, which is to say, it's not just random gibberish.
Further, its second-order entropy – this is a measure of the relative chaos or internal informational relations of the text – do appear language-like from a statistical point of view, though it's actually much lower in terms of its entropic fingerprint than one might expect. Further, the words of the text seem to follow a roughly binomial distribution. With such a distribution and interview markers, the going theory for a while there was what's called the Chinese theory: that Voynich was in an East Asian language written in a Latin-like script, though not indicating the tones. This would be used to explain why there are so many very strange repetitions found in the text, sometimes the same word repeating over and over again in one line. Lacking tone signatures, the apparent repetitions would be different words in pronunciation, but not in transcription.
Though more recent statistical studies have revealed that while some natural languages do have some of the traits displayed by Voynichese, none of them have all or even enough of them to prove to be really viable candidates to be the language. Polynesian languages, for instance, have similar entropy signatures. Chinese and Vietnamese have binomial syllable distributions, and various Gno-native languages might mirror the pre-post and infixes of apparent bases in Voynichese. But ultimately, the exceptions found in Voynichese to any other natural language make it highly unlikely that the text is even an unknown, even exotic, natural language. What we're looking at is not a natural language written in an otherwise unknown alphabet.
Further, if the text were an inciphered European language, it would be highly unlikely that it would have surpassed the cryptological horizon of its day, the mid-15th century. While homophonic and polyalphabetic ciphers did exist at that time, such encryption is easily defeated using contemporary cryptoanalysis. Even the Vigenère cipher, developed in the 1550s following work by Trithemius and de la Porta, was considered indefeasible until it was defeated in 1863. The most complicated cipher of the Middle Ages was defeated in 1863.
Indeed, the very length of the Voynich militates against a very strong encryption technique, and the fluency of the writing of the text either argues that the Voynich is not being encrypted on the fly, or that it's a fair copy of the now-lost preliminary encryption crib sheet. In other words, even the strongest medieval encryption was broken by the 19th century, using pre-machine, much less pre-computational aided analysis. Though some ciphers, especially brief ones, do indeed elude analysis. For instance, the Zodiac Killer's 340 cipher remained unbroken for almost 50 years, mostly because of its brevity and frankly, because of the encoding errors made by the Zodiac Killer in the encryption process.
However, the cipher is to be found in an actual army manual of the time, and with some supercomputers in Australia running the 650,000 iteration, then a really ingenious hunch that the encryption ran along the diagonals, the code was eventually broken just recently, and it provided us another glimpse into the nightmarish mind of that killer.
Of course, the text could also be a constructed language. Dozens of invented alphabets survive from that time period, and small attempts at constructed languages do survive from the Middle Ages. Hildegard of Bingen's Lingua Ignota contains a varied vocabulary of about a thousand words, though these are primarily meant to be placed within an otherwise medieval Latin syntax and ecclesiastical ecosystem. That's to say, she developed a great many words, although almost all of them are nouns, but expected them to function within a larger Latin grammatical system, as is evidenced by some of her surviving songs, which do contain some of her Lingua Ignota.
Further, and closer to the time period of the Voynich production, there's the infamous case of John Dee and Edward Kelley's conversations with angels, in which two statistically distinguished linguistic systems were produced. The first is mostly a set of untranslated text of about 2,800 words, and the second, a group of psalm-like invocations with roughly a thousand words, though it's really close to 250 if you don't count all the repetitions. It's also worth noting that both Dee and Kelley were involved in the court of Rudolph II, but there's actually no evidence of them interacting with the manuscript in any way. And recall that the carbon dating of at least the vellum dates before either of their births, 150 years before it ever shows up in the court of Rudolph II. And further, neither the Enochian language nor the Enochian script are even remotely similar to Voynichese in either appearance or statistical properties, though the sample set of Enochian is so small, it's hard to get good data on exactly what its entropy looks like.
Again, it seems unlikely that such a language could be constructed given medieval understandings of linguistics. And further, even constructed languages tend to analytically appear much like their natural language counterparts, and they do. Voynichese, for instance, I know Klingon and Tolkien's Elvish are pretty exotic languages in a lot of ways, but from an informational and statistical perspective, they behave well within the tolerances of other natural languages.
With the past 20 or so years of profound analysis of the Voynich, I've increasingly come to the position that the text is neither an exotic natural language, nor an encrypted natural language, or a constructed language. Rather, I suspect that Voynichese was produced by a quasi-stochastic method of unknown origin as part of a medieval hoax. By extension, I'm skeptical that the Voynichese text contains any information beyond the remnants of the very process by which it was produced.
Why? Why would I believe that? The first is, well, admittedly the weakest argument: the cryptological horizon for the medieval production of the Voynich is sufficiently low, and the sample size for analysis – the length of the text – is sufficiently large, that contemporary computational, statistical, and cryptological analysis should easily have defeated it by now. It should have defeated any medieval, and it has defeated every medieval cipher up until now. On the other hand, what such analysis has revealed are admittedly strange statistical anomalies, explicable neither in linguistic or cryptological expectations. And more on those in a minute.
But just to start with, a material analysis of the book itself. Despite the appearance of being lavishly illustrated and written in an elegant hand – these are claims that you often see made about the quality of the illustrations or the quality of the scribal arts – that's far from the case. With the exception of some of the bifolios and fold-outs and the draftsmanship of the astrological sections, the illustrations on the Voynich are crude, even by medieval standards. The pigments are often applied haphazardly at best, and the flora and herbal sections are drawn so poorly or just frankly made up that identifications of most of the plants there have proven difficult, if not impossible.
Further, because the script is quill-written and follows a pretty clear ductus – it's roughly a French or Italian ductus of the 15th century – it strikes our modern eyes as kind of beautiful, but truth be told, the script is very plain by contemporary standards. The ductus of the script represents the minimum amount of strokes to compose most of the letters. It doesn't even include serifs or any kind of calligraphic flourishes, really. And aside from a couple of scribbly ascenders and loops, the script lacks even the most basic rudiments of medieval scribal art, such as lubrication or the presence of colored versals, much less illumination. Neither the inks nor the pigments used in the Voynich were especially precious, otherwise extraordinary in any way. The manuscript isn't even scored to allow for a steady, even flow of text on a line. Of course, the illustrations of the text came first, but that's true of many medieval herbals. They simply scored the lines around the illustrations.
Surely the most expensive aspect of the production of the Voynich is the amount of vellum required, roughly about 14 to 15 calves, especially for those those cool fold-outs. But even here, the quality of the vellum is middling. Folios are cut irregularly with odd variations in thickness, even. It's as if they went to the bargain bin of the vellum factory to get some of this stuff. Even Voynich, who had seen plenty of medieval manuscripts in his time, considered this manuscript to be an ugly duckling. While the original binding is lost, the current binding is far from extraordinary and is typical for books of the time. All in all, the Voynich, as a medieval book, isn't terribly remarkable in any way, aside from the fact that no one knows what it says. It's probably the work of between two and five scribe-illustrators, which, if they worked regular hours, could have produced this text in less than a season, much less if that work were further distributed.
In fact, I do calligraphy, and I've copied out sections of the manuscript by hand, and I suspect that a single person working eight hours a day could probably prepare some version of the manuscript in roughly four months, taking weekends off and working during the daylight days of the summer. Overall, the text would have been a financial and labor investment to be sure, but one that could have been engineered precisely to pay off. Yet, it would have taken some investment, but I think that investment would have paid good dividends.
But I think the strongest evidence that this is a hoax isn't to be found primarily in the middling production quality of the book, but in the text of the Voynich Manuscript itself. Now, it's long been noticed that the Voynichese text is highly repetitive, with some strings occurring multiple times per page and often multiple times in a single row of text. The three most statistically significant of these strings are "dane," "ol," and "chetty." I'll be pronouncing them using the EVA system of mapping them out.
Further, it was noticed by Currier back in the day that there is a bimodal distribution of words in Voynich, with "dane" occurring the most overall, but "chetty" following it in a very specific distribution. Though the sections dominated by the appearance of "chetty"-like words do not contain much in the way of the vocabulary from the sections where "dane" dominates and "chetty" is missing. This is pretty stark when you see the distribution. This bimodal distribution became known as Currier A: the "dane" heavy without "chetty," and Currier B: "chetty" heavy and its mutations, although "dane" is still in there as well, to serve Currier A and Currier B languages respectively.
Other statistical observations in this analysis have also come to light. For instance, on a given page, most of the words on that page will only differ in one respect from the three most commonly occurring words on that page. That's really weird. In some of the Currier A sections, the keyword "dane" can occupy up to two-thirds of the most frequent strings on a given page, when a difference of just one glyph is accounted for. In the entire manuscript, something truly remarkable appears. The Currier bimodal distribution is accounted for, but a slow transition from "dane," "ol," and "chetty" appears, with the longest distance between any mutation just being 21 steps. Most are much fewer, and the grouping is shockingly tight, showing just how tightly networked Voynichese words are, just how similar Voynichese is from one word to another.
A recent study has shown that with only one glyph shift, nearly 85% of Voynichese words can be shown to be set into one mutational network, with the central nodes in that shift being the most common Voynichese words: "dane," "ol," and "chetty." These two statistical anomalies are highly unusual for both a natural language. In fact, such repetition is unheard of, even for tonal languages where the tone marks are missing, and for encrypted text, unless there are multiple systems of encryption at work underneath the text we're seeing. But then the boundaries between the internal statistical synoptical relations over the course of the text should look much more stark if different encryption systems are being used. Even a slight difference in the encryption schemes wouldn't likely produce such a network marked by nutritional transition.
Thus, Currier A and B are probably not different languages or encryption systems, but a transition in the text toward a varying priority of word use, either for technical reasons – the text is just shifting from specialized language to specialized language – or for unknown reasons. Perhaps, and I think that is what it is, as an artifact of the very method that was used to produce Voynichese in the first place. But the difference of Currier A and Currier B doesn't track cleanly in terms of shifts from technical field to technical field. There are herbal sections in both Currier A and B, and even the B language difference in relation to other B sections. So, what's going on?
As I've said, I've come to think that we're seeing a quasi-stochastic algorithm-like method being used to generate the text. Not totally random, but not totally natural language, nor the application of a mathematical encryption cycle either. I'm pretty convinced at this point by an article written by Tim and Skinner in 2019 that posits just such a model for producing Voynichese-like text. Like this text here, the tokens "ain," "ol," and "dy" are mutated with those refutation rules, which slightly alter the root by adding or changing, often just a single glyph, as we saw earlier, adding or removing prefixes, which are well-known in the history of the Voynich Manuscript, or combining source words to create new ones.
Further, new pages are seeded by the previous ones, with the mutations unfolding page after page. Even new words seem to follow aesthetic considerations, such as when a word ends in "y," which is quite common, it is four times more likely to be followed up by a word starting with "q" or "o." Further, because of the simplicity of the ductus of Voynichese, making changes to prevent the text from being overly repetitive is often just a flick of the quill away. Indeed, the choice of the seed text not only has ramifications for that page, but it also seems to explain the long-term shift in the manuscript between Currier B and A languages. Once a significantly new mutation is introduced, or a new scribe perhaps fancies a new set of keywords, it's going to be repeated in various mutated and non-mutated forms until something breaks that cycle, like a new scribe, or even an end of the copying shift that day.
The nature of those selection pressures could be anything from whimsy to aesthetics to boredom, carelessness, or just more or less following the rules that were developed for generating the texts of Voynichese in the first place. With a very simple set of rules, Tim and Skinner have been able to reproduce a rather simple algorithm that produces Voynichese-like text that satisfies Zipf's laws, has a very similar entropy signature as the original text, it's matched that odd binomial word length distributions, etc.
Of course, the writer of the Voynich, or the writers of the Voynich, weren't following a strict algorithm, but I think this has the features of explaining the apparent fluency of the Voynich scribe and the overall lack of obvious errors or corrections. The quasi-stochastic algorithm, once you have a string of seed text on a page, is very easy to output. And because there's no meaning, you obviously don't make any mistakes. I've copied out the text like this, just sort of mutating it on my own, and it flows. Further, it explains just how the text may drift between statistically significant words dominating certain sections, and finally, it gives the scribes enough simple rules which, when iterated, produce tantalizing language-like text.
At this point, after working on the Voynich myself as an admitted amateur for 20 years, I strongly suspect that a quasi-stochastic algorithm-like the one being described in this paper by Tim and Skinner was used to produce the Voynich Manuscript. I also think the application of the algorithm is stochastic enough that it will not and cannot be completely incoherently described. There were simply enough rules to create a coherent enough looking text while allowing medieval people, allowing medieval scribes, enough latitude to create as they went, at least to some degree, within certain medieval Voynichese tolerances. Also, and lastly, I think the text might also contain hidden messages. It's certainly possible, but I think it's mostly noise, and I think that even if there are hidden messages, they're probably something like Easter eggs rather than the point of the production of the overall volume as we have it.
But why would anyone do this? Why create such a time-consuming and expensive book of nonsense? Guessing what motivates people to do things, anything, especially strange things like create elaborate hoaxes, is well, endlessly complicated, given just how different medieval subjectivity is from modern subjectivity. For instance, I can imagine a modern person, a postmodern person, creating a hoax like this as part of some bizarre notion of artistic self-expression, whatever that is. But it's hard for me to imagine a medieval equivalent. Medieval people just didn't really seem to think like that. Their subjectivity was frankly very different than ours.
The most reasonable explanation, to my mind at least, is the creation of a hoax for the same primary reasons anyone might do that at any time: to acquire social and financial capital. While the book would have been a relatively expensive hoax, most hoaxes that pay off do require serious time investment and financial investment. You can think about the recent Joash Tablets, the First Temple Ivory Pomegranate, or the Hobby Lobby Dead Sea Scrolls, which were written on shoes from Hadrian's Wall. With a team of two to five scribe-illustrators and a financial backer for securing the parchment, the pigments, and the binding, which should actually represent only about 30 to 40% of the overall investment, if the scribes are willing to work for a share of the breakdown, one could imagine this project could produce a mysterious volume that would fetch a pretty penny.
In fact, we know it did. When the Voynich enters history, we learn from a letter written by Johannes Marcus Marci, yep, Marcus Marci, like Marky Mark and the Funky Bunch, he sent a letter to the polymath Athanasius Kircher in 1665 that the Holy Roman Emperor acquired the text for 600 ducats. Now, a ducat was about 3.5 grams of pure gold. In the fall of 2022, when I'm making this episode, that's worth about $113,400, just in the value of the bullion. Now, getting exact values for medieval currency is notoriously difficult, but the highest skilled guild-backed workers earned about 450 ducats a year around 1500. Thus, we can argue that when the Voynich was sold into the court of the Holy Roman Emperor, it fetched nearly the value of a year and a half, a year and a half worth of highly skilled guild-backed labor. I suspect that if the manuscript fetched even half that value when it entered the book market a century and a half prior to entering the court of Rudolph II, it still would have enjoyed a very handsome profit.
Thus, given the preponderance of this evidence from the middling quality of the Voynich as a work of scribal arts, to the statistical observations that indicate a quasi-algorithmic method for producing the language-like text rather than some super complex incitement process, and the financial incentive for producing just such a work, my best guess is that this document, the most mysterious manuscript in the world, the Voynich Manuscript, is probably, in my opinion, a late medieval hoax.
Of course, I still hold out hope for the possibility that it does hide some secrets, though admittedly, those secrets would probably not much outstrip the knowledge of the mid-15th century world. Despite people seeing everything in this document from cellular life to spiral galaxies and reptile, I don't know. I still hold out hope enough that I still take out my dog-eared copy that's heavily annotated with my own notes. I take it down to my local pub, I grab a boilermaker, hoping that menthol and lubricant will help me to gaze into the curious pages, hoping that I notice some pattern or crib, or that some of the weird books that I've read will give me some special insight into what all those stains and "ols" and "chetty"s – what they mean. And maybe they will make some final sense. But at this point, the best theory that I've seen is the one that I've outlined in this episode. And of course, the best defeat of that theory would be it'd be a decryption. So here's to hoping I'm wrong, despite what I may believe at this juncture, and my long-standing relationship with this ugly duckling of a manuscript mystery. Until next time, I'm Dr. Justin Sledge, and thank you for watching Esoterica, where we explore the arcane in history, philosophy, and religion.